Laser micromachining processes include, for example, semiconductor memory link processing, material trimming, wafer/panel scribing, wafer/panel dicing, and via drilling. Generally, laser micromachining may use laser pulses having wavelengths of, for example, 1.3 μm, 1.064 μm, 1.053 μm, or 1.047 μm, and their harmonics, and pulse widths in a nanosecond range (e.g., between a few nanoseconds and approximately 200 nanoseconds), depending on the particular materials and target structure to be processed, the laser sources used, and the pulse repetition rates used. Laser micromachining processes such as wafer dicing, for example, may use mode-locked laser pulses with pulse widths that are less than approximately 1 nanosecond and pulse repetition rates that are between a few Hz and approximately 300 kHz or higher.
After manufacture of a semiconductor memory array chip is complete, integrated circuit (IC) patterns on an exposed surface of the chip are generally sealed with an electrically insulating layer of passivating material. Typical passivating materials include resins or thermoplastic polymers such as, for example, polyimide, SiO2, or SiN. The purpose of this final “passivation” layer is to prevent the surface of the chip from reacting chemically with ambient moisture, to protect the surface from environmental particulates, and to absorb mechanical stress. Following passivation, the chip is mounted in an electronic package embedded with metal interconnects that allow probing and functional testing of the memory cells. When one of many memory cells is determined to be faulty, the cell is disabled by severing the conductive interconnects, or wires, linking that cell to its neighbors in the array. Wires to one of a plurality of “redundant” memory cells may also be severed to activate the redundant memory cell for replacement of the faulty cell. Disabling and/or activating individual memory cells by “link processing” or “link blowing” is accomplished by laser micromachining equipment that is capable of directing laser beam energy so as to selectively remove the link material in a highly localized region without imparting damage to the materials adjacent to, below, or above the target. Selectively processing a designated link may be achieved by varying the laser beam wavelength, spot size, pulse repetition rate, pulse shape, or other spatial or temporal beam parameters that influence the interaction of the target with the delivered laser energy.
Laser micromachining processes that entail post-processing of electrically conductive links in memory arrays or other types of IC chips use sharp pulses with a fast rising front edge (e.g., with a 0.1 to 10 nanosecond rise time) to achieve desired quality, yield, and reliability. To cleanly sever a link, the laser pulse penetrates the overlying passivation layer before cutting through the metal interconnect. The rising edge of a typical pulse from an existing solid-state laser varies with pulse width. Use of a traditional Gaussian-shaped laser pulse having a 5-20 nanosecond pulse width and a sloped, gradually rising front edge in link processing tends to cause an “over crater” in the passivation layer, especially if its thickness is too large or is uneven. Over cratering reduces the reliability of IC chips.
Rupture behavior of overlying passivation layers has been well analyzed by Yunlong Sun in his PhD dissertation entitled, “Laser processing optimization of semiconductor based devices” (Oregon Graduate Institute, 1997). Because passivation layer thickness is an important parameter, the optimal thickness of a particular passivation layer material may be determined by simulations based on Sun's analysis. Difficulty in maintaining wafer-level process control of the passivation layer during IC fabrication may result in non-optimal thickness and poor cross-wafer or wafer-to-wafer thickness uniformity. Therefore, optimizing characteristics of laser pulses used in post-processing may help to compensate for mis-targeted dimensions and sources of variation in the passivation layer.
U.S. Pat. No. 6,281,471 of Smart proposes using substantially square-shaped laser pulses for link processing. Such a sharp-edged pulse may be generated by coupling a master oscillator laser with a fiber amplifier (MOPA). This low power master oscillator employs a diode laser that is capable of generating a square-shaped pulse with a fast rise time. On the other hand, U.S. Pat. No. 7,348,516 of Yunlong Sun et al., which patent is assigned to the assignee of this patent application, states that, despite a vertical rising edge, a substantially square-shaped laser pulse is not the best laser pulse shape for link processing. Instead, Sun, et al. describes use of a specially tailored laser pulse shape that, in one embodiment, resembles a chair, with a fast rising peak or multiple peaks to most effectively process links, followed by a drop-off in signal strength that remains relatively flat at a lower power level before shutting off. Such a tailored laser pulse, with high peak power but low average power, has been successfully generated by what is called pulse slicing technology, which can be implemented by either electro-optical modulation (EOM) or acousto-optical modulation (AOM). For example, a conventional active Q-switched solid-state laser provides nanosecond seed pulses with high intensity and high pulse energy, and then a light-loop slicing device transforms a standard laser pulse into a desired tailored pulse shape.
U.S. patent application Ser. No. 12/057,264, of Xiaoyuan Peng et al., which application is assigned to the assignee of the present patent application, teaches a light-loop slicing scheme implemented, for example, in an ultraviolet (UV) laser system for semiconductor link processing. Alternatively, a specially tailored laser pulse may be generated by a MOPA (Master Oscillator, Power Amplifier) that employs a gain fiber as the power amplifier. Using a MOPA is advantageous in that it constitutes a stable signal source at a specified constant frequency.
U.S. Patent Application Publication No. 2006/0159138 of Pascal Deladurantaye describes a shaped-pulse laser in which two modulators shape a continuous wave (CW) light beam to generate various shaped pulses. However, generating a pulsed laser from a CW light beam is fairly inefficient, and thus requires more amplification. Because such a low peak-power signal may be influenced by noise, which causes pulse-to-pulse instability, the two modulators are preferably synchronized to maintain pulse stability and energy stability, thereby adding further complexity and cost.
The above systems and methods generally use laser pulses with pulse widths in the nanosecond range. However, laser pulses with pulse widths in the nanosecond range have disadvantages. As has been discussed in detail by Yunlong Sun, “Laser Processing Optimization for Semiconductor Based Devices” (unpublished doctoral thesis, Oregon Graduate Institute of Science and Technology, 1997), conventional laser link processing with nanosecond pulse width may rely on heating, melting, and evaporating the link, and creating a mechanical stress build-up to explosively open the overlying passivation layer with a single laser pulse. Such a conventional link processing laser pulse creates a large heat affected zone (HAZ) that could deteriorate the quality of the device that includes the severed link. For example, when the link is relatively thick or the link material is too reflective to absorb an adequate amount of the laser pulse energy, more energy per laser pulse is used to sever the link. Increased laser pulse energy increases the damage risk to the IC chip, including irregular or over sized opening in the overlying passivation layer, cracking in the underlying passivation layer, damage to the neighboring link structure and damage to the silicon (Si) substrate. However, using laser pulse energy within a risk-free range on thick links often results in incomplete link severing.
Thus, investigations have been performed for using ultrafast lasers (either picosecond or femtosecond lasers) to process semiconductor materials such as links in IC chips. However, the high peak power of a single ultrafast pulse may easily damage the underlying Si substrate, which is unacceptable in many applications. One solution to the problem of high peak power substrate damage caused by ultrafast lasers is to use a burst or train of ultrafast pulses with smaller peak powers. A pulse train with low peak intensity also has the effect of producing a smaller effective spot size in the material. A problem with using a train of ultrafast pulses is that many commercially available ultrafast lasers that use a pulse picker have pulse repetition rates in the 1 kHz to 200 kHz range. Without the pulse picker, a mode-locked laser runs at a fixed repetition rate that is typically in the tens of megahertz range. Such a repetition rate may be difficult to be directly applied to links because stage movement, typically approximately 400 mm/s, is too slow to move the “next” laser pulse to the “next” link within the laser pulse interval time of a few tens of nanoseconds.
U.S. Pat. No. 6,574,250 issued to Yunlong Sun et al., which patent is assigned to the assignee of the present patent application, uses bursts of ultrashort laser pulses to sever conductive links. The pulse width of each laser pulse within a burst may be between 25 picoseconds and 100 femtoseconds. U.S. Patent Application Publication No. 2007/0199927, of Bo Gu et al., uses a laser with at least one pulse having a pulse duration in a range between approximately 10 picoseconds and less than approximately 1 nanosecond. Achim Nebel et al. from Lumera Laser GmbH have demonstrated a passively mode-locked laser that uses digital timing control to generate sequences or groups of pulses. See, “Generation of Tailored Picosecond-Pulse-Trains for Micro-Machining,” Photonics West 2006, LASE Conference: Commercial and Biomedical Applications of Ultrafast Lasers VI Paper No. 6108-37. The system taught by Achim Nebel et al. is based on a “double-switch” scheme generated by high-voltage electro-optical (EO) pulse-picker that drives a voltage passing a half wave of a Pockels cell and generates two HV pulses in one cycle. The delay time between groups of pulses is changeable.